This thesis describes the effort of adapting potential field methods towards man-aging groups of units in real-time strategy game like environments. The focus is ondiscovering the suitability for this technology to create realistic group behaviourand whether it can be adapted to use learning for this purpose.A flexible simulation system for units that are controlled by potential fieldmethods in environments similar to those found in real-time strategy games isdeveloped. This is then used to conduct experiments where different types ofgroup behaviour is sought after. Stochastic optimization techniques are appliedin an attempt to create and optimize desired group behaviour. The results arediscussed and future directions for this research are proposed.